Making person search enjoy the merits of person re-identification

Abstract

Person search is an extended task of person re-identification (Re-ID). However, most existing one-step person search works do not study how to employ existing Re-ID models to improve the one-step person search. To address this issue, we propose a Teacher-guided Disentangling Network (TDN) to make the one-step person search enjoy the merits of existing Re-ID research. The proposed TDN can significantly boost person search performance by transferring the advanced person Re-ID knowledge to the person search model. In the proposed TDN, for better knowledge transfer from the Re-ID teacher model to the one-step person search model, we design a new one-step person search base framework by partially disentangling the two subtasks. Besides, we propose a Knowledge Transfer Bridge module to bridge the scale gap caused by different input formats between the Re-ID model and the one-step person search model. Moreover, we also propose a Ranking with Context Persons strategy to exploit the context information in panoramic images for better ranking. Experiments on two public person search datasets demonstrate the favorable performance of the proposed method.